Title: Machine learning in Raman-based cancer diagnosis
Abstract:
Cancer is undoubtedly one of the most prevalent and lethal diseases in this day and age. So, early diagnosis and treatment of this disease are of paramount importance, drawing the attention of researchers and scientists worldwide. Among various methods of diagnosis, Raman spectroscopy as an optical, non-destructive, and sensitive method has shown promising results in recent years, mainly thanks to machine learning (ML) and artificial intelligence (AI) methods in extracting information from the complicated Raman spectra.
In this presentation, we first get familiar with the specific characteristics of Raman spectra, especially in cancer studies, including the dimensions, interfering signals, and the effect of the measuring system on them and the preprocessing steps. Secondly, there is a discussion over the most relevant and pertinent machine learning methods, including conventional and more modern techniques such as deep and convolutional neural networks, and their pros and cons. Lastly, we deal with the prospect of ML and AI in this field and the limitations and challenges.
It is noteworthy that we will mainly focus on spontaneous single-spot Raman spectroscopy rather than Raman Imaging or other modalities, while we will have a quick look at the differences at the beginning.
I hope this presentation will help researchers pave the way for Raman spectroscopy's translation to oncology clinics, easing the pain of cancer patients.
Audience Take Away Notes:
- The challenges of Raman spectra in terms of data science
- The strengths and weaknesses of conventional ML methods in Raman-based cancer diagnosis
- The pros and cons of using deep learning in the field
- The currently obtained results in both areas (conventional ML and deep learning) according to the literature
- How to pick ML methods based on the study aims and data